Investigating the Impact of Dataset Variants on Heart Disease Prediction Using Supervised Learning Algorithms

In today's modern world due to lifestyle or ancestry, uncertain death or diagnosis of heart diseases have increased significantly not only among senior citizens but also among youths. The early detection of heart disease can help to prevent the risk of one's life. In this paper, authors have compared several machine learning algorithms such as Logistic Regression (LR), Random forest (RF), K-Nearest Neighbour (KNN), Naive Bayes (NB), Support Vector Machine (SVM) and applied these machine learning algorithms on three variants of the dataset i.e. original, scaled and resampled to understand the impact of these variants on prediction system result such that an enhanced and optimal model can be considered or brought into use based on their performance metrics. The dataset consists of 4240 instances and 16 attributes that went through various stages of data preprocessing to handle the anomalies present in the dataset and further where utilised for final training and testing on machine learning models. The authors produced an enhanced performance with accuracy of 85% of Logistic Regression, Support Vector Machine and Random Forest models. Along with accuracy of 82% and 83 % of Naive Bayes and K-Nearest Neighbour models respectively on the original data. All these accuracies of machine learning algorithms were further compared with the accuracies of scaled and resampled dataset on similar algorithms to analyse and understand the impact of dataset variants. The outcome of the research paper encapsulates the concept of efficient working of Machine Learning Algorithms/Models that will help in early detection and prevention of Heart/Cardiovascular diseases.